arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1911.03030cs.LGstat.ML
收藏

从机器学习模型中进行认证数据移除

Certified Data Removal from Machine Learning Models

Cornell University(康奈尔大学);Facebook AI Research(Facebook AI研究院)

Chuan Guo, Tom Goldstein, Awni Hannun, Laurens van der Maaten

AI总结:

针对从训练后机器学习模型中删除数据的需求,本文定义认证移除保证,并为线性分类器提出可证明不可区分的移除机制,且实验验证其实用场景。

摘要:

良好的数据管理要求应数据所有者的请求移除数据。这引出了一个问题:一个隐式存储其训练数据信息的、经过训练的机器学习模型,是否以及如何应受到此类移除请求的影响?是否有可能从机器学习模型中“移除”数据?我们通过定义认证移除来研究这一问题:这是一项非常强的理论保证,即被移除数据的模型无法与一个从一开始就从未观察过该数据的模型区分开来。我们为线性分类器开发了一种认证移除机制,并通过实验研究了该机制在哪些学习设置中具有实用性。

英文摘要:

Good data stewardship requires removal of data at the request of the data's owner. This raises the question if and how a trained machine-learning model, which implicitly stores information about its training data, should be affected by such a removal request. Is it possible to "remove" data from a machine-learning model? We study this problem by defining certified removal: a very strong theoretical guarantee that a model from which data is removed cannot be distinguished from a model that never observed the data to begin with. We develop a certified-removal mechanism for linear classifiers and empirically study learning settings in which this mechanism is practical.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑